April 2014 - May 2014
This was a month long final project in a group of two, for Introduction to Image & Video Processing, ECE418.
For this project, the task is class based pixel-wise segmentation and categorization using the single-histogram class model, and random forests. Histogram of visual words, textons, has been successfully used as the feature, for image segmentation and categorization task, to various supervised learning algorithm, such as k-nearest neighbor, or decisions trees. We investigate and implement the single-histogram class models, and random forest algorithm. These two methods are evaluated on the Microsoft research Cambridge object recognition image database. We have achieved a reasonable performance on the task considering the limited running time and computing power we had.
Image below on the left is an illustration of the texton map, where each color represents a texton. The image on the right is a sample output label image from our system.
For implementation details and results please refer to our final report.